CN-121980526-A - Plastic molten state discrimination method and system based on multi-mode feature fusion

CN121980526ACN 121980526 ACN121980526 ACN 121980526ACN-121980526-A

Abstract

The invention provides a plastic melting state judging method and system based on multi-mode feature fusion, and relates to the technical field of state identification, wherein the method comprises the steps of collecting multi-mode data in a plastic melting process; the method comprises the steps of constructing a cross-modal cross-covariant matrix for describing a collaborative change state among various modal data based on the multi-modal data, constructing a reference cross-covariant matrix representing different molten states by utilizing a organ-generated inequality convex lower bound constraint, constructing a divergence feature vector with the length consistent with the number of the molten states by taking matrix-to-matrix divergence between the cross-modal cross-covariant matrix and each reference cross-covariant matrix as a basic feature, inputting the divergence feature vector into a random forest established based on Fisher information and decision diversity gain, outputting posterior probability of each molten state, and outputting a molten state corresponding to the maximum posterior probability as a plastic molten state discrimination result.

Inventors

SUN BAOLIN
JIANG YINGHAN
ZHANG GUOCHEN
QI HONGMIN

Assignees

嘉禾聚能(北京)科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260409

Claims (10)

1. A plastic melting state discriminating method based on multi-mode feature fusion is characterized by comprising the following steps: S1, collecting multi-mode data in a plastic melting process; S2, constructing a cross-mode covariant matrix for describing the covariant change state among the mode data based on the multi-mode data; s3, constructing a reference mutual covariant matrix representing different melting states by utilizing the convex lower bound constraint of the qin inequality; S4, constructing a divergence characteristic vector with the length consistent with the number of the molten states by taking matrix-to-determinant divergences between the cross-modal mutual covariant matrix and each reference mutual covariant matrix as basic characteristics; S5, inputting the divergence eigenvectors into a random forest established based on Fisher information and decision diversity gain, and outputting posterior probability of each molten state; And S6, outputting a molten state corresponding to the maximum posterior probability as a plastic molten state discrimination result.
2. The method for discriminating a molten state of plastic based on multi-modal feature fusion according to claim 1 wherein said multi-modal data includes a sequence of time-synchronized visual images, a sequence of infrared thermal images and a sequence of acoustic emission time-series signals.
3. The method for determining the molten state of plastic based on multi-modal feature fusion according to claim 2, wherein S2 specifically comprises: s201, respectively extracting transient variation of each mode data in the multi-mode data; S202, establishing the cross-modal cross-covariant matrix according to the transient variation, wherein each element in the cross-modal cross-covariant matrix is a normalized outer product matrix between statistical feature vectors corresponding to any two modal data.
4. The method for distinguishing the molten state of the plastic based on the multi-mode feature fusion according to claim 3, wherein the transient variation is specifically the difference between statistical feature vectors at a first moment and a second moment; Wherein the first time and the second time are adjacent times; the statistical feature vector comprises visual image sequence features, infrared thermal image sequence features and acoustic emission time sequence signal sequence features; The visual image sequence features comprise an image RGB mean value, an image RGB variance and an image histogram peak value, the infrared thermal image sequence features comprise a thermal image average temperature, a thermal image maximum temperature and a thermal image temperature standard deviation, and the acoustic emission time sequence signal sequence features comprise a signal main frequency, a signal spectrum centroid and a signal bandwidth.
5. The method for determining molten state of plastic based on multi-modal feature fusion according to claim 1, wherein the molten state includes an unmelted state, a starting molten state and a completely molten state, and the step S3 specifically includes: s301, acquiring a time index set of each molten state in a history labeling multi-mode data sample set; s302, establishing a history cross-modal covariant matrix corresponding to the same molten-state multi-modal data in the time index set; s303, calculating the matrix logarithm of each history cross-modal mutual covariant matrix; s304, sequentially carrying out arithmetic mean value operation and matrix index mapping operation on the calculation result to obtain an initial reference mutual covariant matrix; s305, adding a regular term to the initial reference mutual covariant matrix under the constraint of the convex lower bound constraint of the organ-generated inequality to obtain the reference mutual covariant matrix which is always positive.
6. The method for determining the molten state of plastic based on multi-modal feature fusion according to claim 1, wherein S4 specifically comprises: s401, taking the product of the cross-modal cross-covariant matrix and the reference cross-covariant matrix corresponding to different melting states as input, and calculating the matrix logarithmic determinant divergence of the corresponding melting states; and S402, splicing matrix-to-determinant divergences corresponding to the molten states to obtain the divergence eigenvector.
7. The method for distinguishing the molten state of the plastic based on the multi-modal feature fusion according to claim 1, wherein the random forest comprises an input layer, an integration layer and an output layer which are sequentially connected, the integration layer comprises a plurality of parallel decision trees, and the step S5 specifically comprises: S501, determining a hyper-parameter set of the random forest by combining the Fisher information with the aim of ensuring that each decision tree in the random forest has decision diversity, wherein the hyper-parameter set comprises a time sequence forgetting factor, the number of the decision trees of the random forest and the depth of the decision trees; S502, updating the random forest according to the determined number of the random forest decision trees and the depth of the decision trees; s503, according to the determined time sequence forgetting factor, establishing a time sequence input vector related to the divergence characteristic vector; s504, inputting the time sequence input vector into the updated random forest, and outputting the posterior probability of each molten state.
8. The method for determining a molten state of plastic based on multi-modal feature fusion according to claim 7, wherein S501 specifically comprises: s5011, acquiring training data, wherein the training data comprises a sample to be distinguished and a corresponding molten state label; S5012, encoding different hyper-parameter groups of the random forest into population individuals, and initializing the population individuals; S5013, establishing an adaptability function of the decision diversity gain of the random forest; S5014, inputting the sample to be judged into the random forest, and outputting a molten state prediction label; S5015, calculating the fitness function value of each population individual, and entering a step S5019 when the change amount of the fitness function value is larger than the change amount of a preset fitness function value or the iteration number is larger than the preset iteration number, otherwise, entering a step S5016; s5016, calculating the adaptability gradient of the current population; S5017, establishing an experience Fisher information matrix based on the fitness gradient so as to quantify the sensitivity of each super parameter to the fitness function; S5018, determining a super-parameter group searching direction based on the experience Fisher information matrix, updating the population individuals according to the super-parameter group searching direction, and returning to the step S5014; s5019, outputting the current super parameter set.
9. A plastic molten state discrimination system based on multi-modal feature fusion, comprising: A processor; A memory having stored thereon computer readable instructions which, when executed by the processor, implement the method of plastics melt state discrimination based on multimodal feature fusion of any one of claims 1to 8.
10. A computer-readable storage medium having stored thereon a computer program, which when executed by a processor implements the plastic molten state discrimination method based on multi-modal feature fusion according to any one of claims 1 to 8.

Description

Plastic molten state discrimination method and system based on multi-mode feature fusion Technical Field The invention relates to the technical field of state identification, in particular to a method and a system for discriminating a molten state of plastic based on multi-mode feature fusion. Background The molten state of plastic refers to the different stages of plastic that gradually change from a solid state to a liquid state during heating. Generally, they can be classified into an unmelted state (solid state, closely packed molecular chains), a starting melted state (a part of molecules starts to flow, plastics soften), and a completely melted state (plastics are completely liquefied, and molecular chains are sufficiently moved). Plastic melt state identification is directly related to plastic processing quality and product performance. The process parameters such as injection molding, extrusion and the like can be effectively controlled by accurately judging the molten state, so that defects caused by early or late molding are avoided, meanwhile, the production efficiency and the material utilization rate are improved, the energy consumption and the production cost are reduced, and the mechanical property, the appearance and the durability of the final product are ensured to meet the design requirements. However, in the prior art, the judgment of the molten state of the plastic depends on single-mode data or directly judges the molten state by subjectively setting the heating time length, so that the cooperative change between different physical signals is difficult to capture, the judgment precision is low, the error is large, and the real-time and reliable state monitoring and processing quality control cannot be realized. Disclosure of Invention The invention provides a plastic molten state judging method and a system based on multi-mode feature fusion, which are used for solving the technical problems that in the prior art, the plastic molten state is judged by depending on single-mode data or the molten state is judged directly by subjectively setting heating time length, and the cooperative change between different physical signals is difficult to capture, so that the judgment precision is low, the error is large, and real-time and reliable state monitoring and processing quality control cannot be realized. The technical scheme provided by the embodiment of the invention is as follows: the first aspect of the embodiment of the invention provides a plastic melting state judging method based on multi-mode feature fusion, which comprises the following steps: S1, collecting multi-mode data in a plastic melting process; s2, constructing a cross-mode covariant matrix for describing the covariant change state among the mode data based on the multi-mode data; s3, constructing a reference mutual covariant matrix representing different melting states by utilizing the convex lower bound constraint of the qin inequality; S4, constructing a divergence characteristic vector with the length consistent with the quantity of the molten state by taking matrix-to-determinant divergences between the cross-modal cross-covariant matrix and each reference cross-covariant matrix as basic characteristics; S5, inputting the divergence eigenvector into a random forest established based on Fisher information and decision diversity gain, and outputting the posterior probability of each molten state; And S6, outputting a molten state corresponding to the maximum posterior probability as a plastic molten state discrimination result. The second aspect of the embodiment of the invention provides a plastic melting state discrimination system based on multi-mode feature fusion, which comprises: A processor; And a memory having stored thereon computer readable instructions which, when executed by the processor, implement the method for determining molten state of plastic based on multi-modal feature fusion according to the first aspect. A third aspect of the embodiments of the present invention provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the method for determining a molten state of plastic based on multimodal feature fusion as described in the first aspect. The technical scheme provided by the embodiment of the invention has the beneficial effects that at least: In the invention, the limitation of a single mode is broken through by capturing the collaborative dynamic change of the multi-mode data in the plastic melting process through the cross-mode mutual covariant matrix. The method comprises the steps of constructing a molten state reference matrix by using a piano generation inequality, essentially establishing statistical prototype templates of different molten stages, quantifying the difference between real-time data and each state prototype by using matrix-to-determinant divergence, and converting a high-dimensional matrix relation int